A memory problem with ecp basis sets

[FreemanTheMaverick]

In a previous issue(#251 (comment)), I reported a memory problem with ecp basis sets of iodine. Although it has not been solved, I managed to achieve my goal with some other methods supported by other softwares. But now I am caught in a scenario where CASPT2 analytical gradient of an iodine-containing molecule is necessary.

To see whether the problem occurs only in iodine, I tested on some other elements with lanl2dz, which offers ecp for elements from Na. The result is that the problem disappears in Ca-Ti and Cu but shows up in Na-Si, K, Zn, Br and I. According to the limited test, I guess the problem may well occurs in elements other than Ca-Cu.

Bagel does not provide many ecp basis sets by itself and the script convert.py does not seem to be able to deal with ecp basis sets, so I wrote a python script to convert gaussian's basis set files to Bagel's format. Here I attach lanl2dz-ecp+.json generated by my script. I will also attach the script once the problem is eliminated, since the script will never be useful until then.

lanl2dz-ecp+.json.txt

[shiozaki]

Just to note that ECP code in the open-source version of BAGEL is rather preliminary and not fully functional. Those are rewritten in the QSimulate version. https://qsimulate.com/academic A precompiled version is available for a fee (the website emphasizes QM/MM simulation but all of the BAGEL QM features + some additional ones are available).